A Markov Chains based transition matrices approach to forecasting airline seat demand

The key to the billions of dollars made yearly in the hospitality industry, particularly the airline and hotel industries, lies in its yield and revenue management systems. Yet very few people outside of the industry realize the complexity of those systems and what drives it. Any yield and revenue management system is only as good as its [over]booking engine. And if booking is to be considered the engine to those systems, then the forecasting module is the fuel that powers it.;The strength of a forecast is always measured by its accuracy but that can only be determined after the fact. Forecasts for airlines and hotels are subject to a myriad of factors including competitor actions, economic environments and natural disasters. As a result, forecasting in general and more specifically for the hospitality industries is extremely complicated. This dissertation will investigate the Markov Chains approach to forecasting as a simpler solution.;The literature review within this dissertation provides an illustration of the various forecasting methods used. A Markovian based transition matrices approach is then proposed as an alternative method to improve demand forecasting. Its usage and performance criteria are then outlined and a methodology for testing its effectiveness under various simulated environmental conditions is described. A select few of these forecasting methods are compared to the proposed Markov Chain transitional matrices approach in terms of their forecasting accuracy.;Results indicate that a Markov Chains based transition matrices approach does indeed offer an improvement over current selected methods under certain conditions. Analysis also shows that capacity is irrelevant to forecasting accuracy and that the multiplicative pickup methods were the worst performers. Further investigation also reveals that an observation point scheme utilizing data up to 60 days prior to departure was best; and contrary to common statistical principles, a disaggregated fare class structure was the most efficient. In a self-fulfilling prophecy, it was also shown that the forecasting methods were more accurate for censored data.